scikit-learn

Node.js wrapper of scikit-learn

Installation

npm install scikit-learn

Usage

var scikit = require('scikit-learn')

Example

var inspect = require('inspect-stream');

var arrayify = require('arrayify-merge.s');
var slice    = require('slice-flow.s');

var scikit = require('scikit-learn');

var features = scikit.dataset('load_digits.data'); //stream of features
var labels   = scikit.dataset('load_digits.target'); //stream of labels

// arrayify is transform stream that turns two input streams
// into one stream by wraping packets of inputs in array.
// So trainingSet outputs arrays [<feature>, <label>]
var trainingSet = arrayify();
features.pipe(trainingSet);
labels.pipe(trainingSet);

var clf = scikit.svm('SVC', {
  gamma: 0.001,
  C:     100
});

trainingSet
  .pipe(slice([0, -1])) //passes all packets except last one
  .pipe(clf)
  .on('error', function (err) {
    console.log(err);
  })
  .on('end', function () {
    // now we have trained model

    var predict = clf.predict();
    var features = scikit.dataset('load_digits.data');
    features.pipe(slice(-1)) //passes only last packet
      .pipe(predict)
      .pipe(inspect());
  });

API

scikit.dataset(name, options)

• name String Name of method of sklearn.datasets on python side concatenated by dot with name of dataset's subset Ex: 'load_digits.target'
• options Object Options of method

Returns readable stream of dataset

Fit streams

All fit streams are transform streams that acts like writable. So you must listen on end event instead of finish to be sure that training finished

Accepts flow of arrays like [features, label] where 'features' is array of features and label is... label

Also fit stream have event 'model' that emits with trained model. Model is Buffer containing pickled object

Fit stream have method predict that returns Predict stream

scikit.svm(name, options)

• name String Name of method of sklearn.svm
• options Object Options for estimator

Predict streams

Predict stream is transform stream that accepts flow of arrays of features and outputs predictions